update tts

app.py

@@ -8,6 +8,7 @@ from nemo.collections.tts.models import MixerTTSModel
 
 spec_generator = FastPitchModel.from_pretrained("tts_en_fastpitch_multispeaker")
 voc_model = HifiGanModel.from_pretrained(model_name="tts_en_hifitts_hifigan_ft_fastpitch")
+voc_model.eval()
 
 def greet(name):
     return "Hello " + name + "!!"
@@ -17,7 +18,7 @@ def generate_tts(text: str, speaker: int = 0):
     parsed = spec_generator.parse(text)
     spectrogram = spec_generator.generate_spectrogram(tokens=parsed, speaker=speaker)
     audio = voc_model.convert_spectrogram_to_audio(spec=spectrogram)
-    return sr, audio
+    return sr, audio.to('cpu').detach().numpy()
 
 def run():
     demo = gr.Interface(